How AI Is Changing the Study of Atomic Vibrations

Thanks to the emergence of new artificial intelligence (AI) tools and applications, the field of materials science is changing rapidly. Because of the current ongoing changes in this industry, a team of researchers from the Massachusetts Institute of Technology and Oak Ridge National Laboratory decided to document them in a recent review article published in Digital Discovery (1).

In their review article, the researchers discuss how new techniques are driving forward significant changes in studying atomic vibrations. To start their review, they first highlight the importance of atomic vibrations in understanding a material’s thermal and mechanical properties (1). They play a decisive role in heat transfer, energy dissipation, and performance in applications ranging from energy harvesting to carbon capture (1). They also discuss why the traditional approaches, which include ab initio computational methods or experimental spectroscopy, fall short.

A digital visualization of a particle wave in space, illustrating quantum mechanics or wave-particle duality with energy waves and cosmic background. Generated with AI. | Image Credit: © soysuwan123 - stock.adobe.com

High-accuracy ab initio calculations can require massive computational power, often stretching supercomputing resources to their limits, the authors wrote. As a result, it is very time-consuming and resource-intensive (1,2). On the experimental side, measuring vibrational behaviors with the necessary precision demands advanced instrumentation, lengthy procedures, and highly skilled operators (1). Progress has not accelerated at a fast enough rate to keep pace with industry needs, which has led to downstream effects in slowing the progress of thermal storage and waste heat recovery (1,3).

However, thanks to AI, researchers are beginning to experiment with how these tools can improve accuracy and address the limitations of traditional methods. By learning from large data sets of previously computed or experimentally measured results, AI models can predict vibrational spectra, lattice dynamics, and molecular motions at rapid speeds (1,4).

There are several other notable advancements in materials science that the researchers address in their study. One of these advancements are the development of machine learning interatomic potentials (MLIPs). MLIPs are designed to allow for accurate simulations of atomic interactions without the heavy computational cost of direct quantum mechanical calculations (1).

Another key advancement in materials science using AI is the development of graph neural networks (GNNs). These are capable of capturing the complex connectivity of atoms within materials (1). As a result, GNNs have the ability to model structure–dynamics relationships more efficiently than traditional algorithms (1).

AI is also being routinely integrated into the standard workflow for analyzing atomic vibrations. For example, AI algorithms can process and interpret spectroscopic data in real time, enabling scientists to adjust experiments immediately (1). The researchers believe that this change can ultimately drive inverse design forward, where researchers can start with a desired property, such as low thermal conductivity, and use AI to design a material that meets that specification from the atomic level upward (1).

The authors acknowledge that for AI to be used effectively, materials scientists need to work on developing software with comprehensive documentation. Doing so would help increase the amount of people that can work with the AI tools effectively (1). This democratization of advanced modeling has the potential to foster cross-disciplinary collaborations and accelerate discoveries in unexpected directions (1).

Currently in materials science, one of the main objectives is to better understand the structure–dynamics–property relationship. By linking a material’s atomic arrangement to its vibrational behavior and, ultimately, to its performance, researchers can design materials more effectively (1). AI allows scientists to explore this relationship in greater detail and at far greater speeds than ever before (1).

As computing power grows, data sets expand, and algorithms evolve, the authors see the field entering a new AI-powered era of materials research. This shift promises not only faster discovery but also smarter, more targeted development of materials for energy efficiency, sustainability, and high-performance applications (1).

References

1. Han, B.; Okabe, R.; Chotrattanapituk, A.; et al. AI-powered Exploration of Molecular Vibrations, Phonons, and Spectroscopy. Digital Discovery 2025, 4, 584–624. DOI: 10.1039/D4DD00353E
2. Makhnev, V. Y.; Kyuberis, A. A.; Zobov, N. F.; et al. High Accuracy ab Initio Calculations of Rotational–Vibrational Levels of the HCN/HNC System. J. Phys. Chem. A 2018, 122 (5), 1326–1343. DOI: 10.1021/acs.jpca.7b10483
3. Ortega-Fernandez, I.; Rodriguez-Aseguinolaza, J. Thermal Energy Storage for Waste Heat Recovery in the Steelworks: The Case Study of the REslag Project. Appl. Energy 2019, 237, 708–719. DOI: 10.1016/j.apenergy.2019.01.007
4. Han, R.; Ketkaew, R.; Luber, S. A Concise Review on Recent Developments of Machine Learning for the Prediction of Vibrational Spectra. J. Phys. Chem. A. 2022, 126 (6). DOI: 10.1021/acs.jpca.1c10417